Update bufw handling in all gpus.

This fixes the softgpu as well, at least.

CMakeLists.txt

@@ -1002,6 +1002,8 @@ add_library(GPU OBJECT
 	GPU/Common/VertexDecoderCommon.h
 	GPU/Common/IndexGenerator.cpp
 	GPU/Common/IndexGenerator.h
+	GPU/Common/TextureDecoder.cpp
+	GPU/Common/TextureDecoder.h
 	GPU/GLES/GLES_GPU.cpp
 	GPU/GLES/GLES_GPU.h
 	GPU/GLES/FragmentShaderGenerator.cpp

GPU/Common/TextureDecoder.cpp

@@ -0,0 +1,20 @@
+// Copyright (c) 2012- PPSSPP Project.
+
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, version 2.0 or later versions.
+
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public License 2.0 for more details.
+
+// A copy of the GPL 2.0 should have been included with the program.
+// If not, see http://www.gnu.org/licenses/
+
+// Official git repository and contact information can be found at
+// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
+
+#include "GPU/Common/TextureDecoder.h"
+
+// TODO: Move some common things into here.

GPU/Common/TextureDecoder.h

@@ -0,0 +1,74 @@
+// Copyright (c) 2012- PPSSPP Project.
+
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, version 2.0 or later versions.
+
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public License 2.0 for more details.
+
+// A copy of the GPL 2.0 should have been included with the program.
+// If not, see http://www.gnu.org/licenses/
+
+// Official git repository and contact information can be found at
+// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
+
+#pragma once
+
+#include "Core/MemMap.h"
+#include "GPU/ge_constants.h"
+#include "GPU/GPUState.h"
+
+static const u8 textureBitsPerPixel[16] = {
+	16,  //GE_TFMT_5650,
+	16,  //GE_TFMT_5551,
+	16,  //GE_TFMT_4444,
+	32,  //GE_TFMT_8888,
+	4,   //GE_TFMT_CLUT4,
+	8,   //GE_TFMT_CLUT8,
+	16,  //GE_TFMT_CLUT16,
+	32,  //GE_TFMT_CLUT32,
+	4,   //GE_TFMT_DXT1,
+	8,   //GE_TFMT_DXT3,
+	8,   //GE_TFMT_DXT5,
+	0,   // INVALID,
+	0,   // INVALID,
+	0,   // INVALID,
+	0,   // INVALID,
+	0,   // INVALID,
+};
+
+// Masks to downalign bufw to 16 bytes, and wrap at 2048.
+static const u32 textureAlignMask16[16] = {
+	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_5650,
+	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_5551,
+	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_4444,
+	0x7FF & ~(((8 * 16) / 32) - 1),  //GE_TFMT_8888,
+	0x7FF & ~(((8 * 16) / 4) - 1),   //GE_TFMT_CLUT4,
+	0x7FF & ~(((8 * 16) / 8) - 1),   //GE_TFMT_CLUT8,
+	0x7FF & ~(((8 * 16) / 16) - 1),  //GE_TFMT_CLUT16,
+	0x7FF & ~(((8 * 16) / 32) - 1),  //GE_TFMT_CLUT32,
+	0x7FF & ~(((8 * 16) / 4) - 1),   //GE_TFMT_DXT1,
+	0x7FF & ~(((8 * 16) / 8) - 1),   //GE_TFMT_DXT3,
+	0x7FF & ~(((8 * 16) / 8) - 1),   //GE_TFMT_DXT5,
+	0,   // INVALID,
+	0,   // INVALID,
+	0,   // INVALID,
+	0,   // INVALID,
+	0,   // INVALID,
+};
+
+static inline u32 GetTextureBufw(int level, u32 texaddr, GETextureFormat format) {
+	// This is a hack to allow for us to draw the huge PPGe texture, which is always in kernel ram.
+	if (texaddr < PSP_GetUserMemoryBase())
+		return gstate.texbufwidth[level] & 0x1FFF;
+
+	u32 bufw = gstate.texbufwidth[level] & textureAlignMask16[format];
+	if (bufw == 0) {
+		// If it's less than 16 bytes, use 16 bytes.
+		bufw = (8 * 16) / textureBitsPerPixel[format];
+	}
+	return bufw;
+}

GPU/Directx9/TextureCacheDX9.cpp

@@ -24,6 +24,7 @@
 #include "GPU/GPUState.h"
 #include "GPU/Directx9/TextureCacheDX9.h"
 #include "GPU/Directx9/FramebufferDX9.h"
+#include "GPU/Common/TextureDecoder.h"
 #include "Core/Config.h"
 
 #include "ext/xxhash.h"
@@ -44,12 +45,6 @@ namespace DX9 {
 #define TEXCACHE_DECIMATION_INTERVAL 13
 
 extern int g_iNumVideos;
-static inline u32 GetLevelBufw(int level, u32 texaddr) {
-	// Special rules for kernel textures (PPGe):
-	if (texaddr < PSP_GetUserMemoryBase())
-		return gstate.texbufwidth[level] & 0x1FFF;
-	return gstate.texbufwidth[level] & 0x7FF;
-}
 
 TextureCacheDX9::TextureCacheDX9() : clearCacheNextFrame_(false), lowMemoryMode_(false), clutBuf_(NULL) {
 	lastBoundTexture = INVALID_TEX;
@@ -407,8 +402,7 @@ inline void DeIndexTexture4Optimal(ClutT *dest, const u32 texaddr, int length, C
 	DeIndexTexture4Optimal(dest, indexed, length, color);
 }
 
-void *TextureCacheDX9::readIndexedTex(int level, u32 texaddr, int bytesPerIndex, u32 dstFmt) {
+void *TextureCacheDX9::ReadIndexedTex(int level, u32 texaddr, int bytesPerIndex, u32 dstFmt, int bufw) {
-	int bufw = GetLevelBufw(level, texaddr);
 	int w = gstate.getTextureWidth(level);
 	int h = gstate.getTextureHeight(level);
 	int length = bufw * h;
@@ -801,25 +795,6 @@ void TextureCacheDX9::StartFrame() {
 	}
 }
 
-static const u8 bitsPerPixel[16] = {
-	16,  //GE_TFMT_5650,
-	16,  //GE_TFMT_5551,
-	16,  //GE_TFMT_4444,
-	32,  //GE_TFMT_8888,
-	4,   //GE_TFMT_CLUT4,
-	8,   //GE_TFMT_CLUT8,
-	16,  //GE_TFMT_CLUT16,
-	32,  //GE_TFMT_CLUT32,
-	4,   //GE_TFMT_DXT1,
-	8,   //GE_TFMT_DXT3,
-	8,   //GE_TFMT_DXT5,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-};
-
 static inline u32 MiniHash(const u32 *ptr) {
 	return ptr[0];
 }
@@ -856,7 +831,7 @@ static inline u32 QuickClutHash(const u8 *clut, u32 bytes) {
 }
 
 static inline u32 QuickTexHash(u32 addr, int bufw, int w, int h, GETextureFormat format) {
-	const u32 sizeInRAM = (bitsPerPixel[format] * bufw * h) / 8;
+	const u32 sizeInRAM = (textureBitsPerPixel[format] * bufw * h) / 8;
 	const u32 *checkp = (const u32 *) Memory::GetPointer(addr);
 	u32 check = 0;
 
@@ -1083,7 +1058,7 @@ void TextureCacheDX9::SetTexture() {
 
 	int w = gstate.getTextureWidth(0);
 	int h = gstate.getTextureHeight(0);
-	int bufw = GetLevelBufw(0, texaddr);
+	int bufw = GetTextureBufw(0, texaddr, format);
 	int maxLevel = ((gstate.texmode >> 16) & 0x7);
 
 	u32 texhash = MiniHash((const u32 *)Memory::GetPointer(texaddr));
@@ -1236,7 +1211,7 @@ void TextureCacheDX9::SetTexture() {
 
 	// This would overestimate the size in many case so we underestimate instead
 	// to avoid excessive clearing caused by cache invalidations.
-	entry->sizeInRAM = (bitsPerPixel[format] * bufw * h / 2) / 8;
+	entry->sizeInRAM = (textureBitsPerPixel[format] * bufw * h / 2) / 8;
 
 	entry->fullhash = fullhash == 0 ? QuickTexHash(texaddr, bufw, w, h, format) : fullhash;
 	entry->cluthash = cluthash;
@@ -1300,7 +1275,7 @@ void *TextureCacheDX9::DecodeTextureLevel(GETextureFormat format, GEPaletteForma
 
 	u32 texaddr = (gstate.texaddr[level] & 0xFFFFF0) | ((gstate.texbufwidth[level] << 8) & 0x0F000000);
 
-	int bufw = GetLevelBufw(level, texaddr);
+	int bufw = GetTextureBufw(level, texaddr, format);
 
 	int w = gstate.getTextureWidth(level);
 	int h = gstate.getTextureHeight(level);
@@ -1310,9 +1285,6 @@ void *TextureCacheDX9::DecodeTextureLevel(GETextureFormat format, GEPaletteForma
 	case GE_TFMT_CLUT4:
 		{
 		dstFmt = getClutDestFormat(clutformat);
-		// Don't allow this to be less than 16 bytes (32 * 4 / 8 = 16.)
-		if (bufw < 32)
-			bufw = 32;
 
 		const bool mipmapShareClut = (gstate.texmode & 0x100) == 0;
 		const int clutSharingOffset = mipmapShareClut ? 0 : level * 16;
@@ -1371,34 +1343,26 @@ void *TextureCacheDX9::DecodeTextureLevel(GETextureFormat format, GEPaletteForma
 		break;
 
 	case GE_TFMT_CLUT8:
-		if (bufw < 8)
-			bufw = 8;
 		dstFmt = getClutDestFormat(gstate.getClutPaletteFormat());
 		texByteAlign = texByteAlignMap[gstate.getClutPaletteFormat()];
-		finalBuf = readIndexedTex(level, texaddr, 1, dstFmt);
+		finalBuf = ReadIndexedTex(level, texaddr, 1, dstFmt, bufw);
 		break;
 
 	case GE_TFMT_CLUT16:
-		if (bufw < 8)
-			bufw = 8;
 		dstFmt = getClutDestFormat(gstate.getClutPaletteFormat());
 		texByteAlign = texByteAlignMap[gstate.getClutPaletteFormat()];
-		finalBuf = readIndexedTex(level, texaddr, 2, dstFmt);
+		finalBuf = ReadIndexedTex(level, texaddr, 2, dstFmt, bufw);
 		break;
 
 	case GE_TFMT_CLUT32:
-		if (bufw < 4)
-			bufw = 4;
 		dstFmt = getClutDestFormat(gstate.getClutPaletteFormat());
 		texByteAlign = texByteAlignMap[gstate.getClutPaletteFormat()];
-		finalBuf = readIndexedTex(level, texaddr, 4, dstFmt);
+		finalBuf = ReadIndexedTex(level, texaddr, 4, dstFmt, bufw);
 		break;
 
 	case GE_TFMT_4444:
 	case GE_TFMT_5551:
 	case GE_TFMT_5650:
-		if (bufw < 8)
-			bufw = 8;
 		if (format == GE_TFMT_4444)
 			dstFmt = D3DFMT_A4R4G4B4;
 		else if (format == GE_TFMT_5551)
@@ -1422,8 +1386,6 @@ void *TextureCacheDX9::DecodeTextureLevel(GETextureFormat format, GEPaletteForma
 		break;
 
 	case GE_TFMT_8888:
-		if (bufw < 4)
-			bufw = 4;
 		dstFmt = D3DFMT_A8R8G8B8;
 		if (!gstate.isTextureSwizzled()) {
 			// Special case: if we don't need to deal with packing, we don't need to copy.

GPU/Directx9/TextureCacheDX9.h

@@ -118,7 +118,7 @@ private:
 
 	void Decimate();  // Run this once per frame to get rid of old textures.
 	void *UnswizzleFromMem(u32 texaddr, u32 bufw, u32 bytesPerPixel, u32 level);
-	void *readIndexedTex(int level, u32 texaddr, int bytesPerIndex, u32 dstFmt);
+	void *ReadIndexedTex(int level, u32 texaddr, int bytesPerIndex, u32 dstFmt, int bufw);
 	void UpdateSamplingParams(TexCacheEntry &entry, bool force);
 	void LoadTextureLevel(TexCacheEntry &entry, int level, bool replaceImages);
 	void *DecodeTextureLevel(GETextureFormat format, GEPaletteFormat clutformat, int level, u32 &texByteAlign, u32 &dstFmt);

GPU/GLES/TextureCache.cpp

@@ -24,6 +24,7 @@
 #include "GPU/GPUState.h"
 #include "GPU/GLES/TextureCache.h"
 #include "GPU/GLES/Framebuffer.h"
+#include "GPU/Common/TextureDecoder.h"
 #include "Core/Config.h"
 
 #include "ext/xxhash.h"
@@ -45,13 +46,6 @@
 
 extern int g_iNumVideos;
 
-static inline u32 GetLevelBufw(int level, u32 texaddr) {
-	// Special rules for kernel textures (PPGe):
-	if (texaddr < PSP_GetUserMemoryBase())
-		return gstate.texbufwidth[level] & 0x1FFF;
-	return gstate.texbufwidth[level] & 0x7FF;
-}
-
 TextureCache::TextureCache() : clearCacheNextFrame_(false), lowMemoryMode_(false), clutBuf_(NULL) {
 	lastBoundTexture = -1;
 	decimationCounter_ = TEXCACHE_DECIMATION_INTERVAL;
@@ -776,45 +770,6 @@ void TextureCache::StartFrame() {
 	}
 }
 
-static const u8 bitsPerPixel[16] = {
-	16,  //GE_TFMT_5650,
-	16,  //GE_TFMT_5551,
-	16,  //GE_TFMT_4444,
-	32,  //GE_TFMT_8888,
-	4,   //GE_TFMT_CLUT4,
-	8,   //GE_TFMT_CLUT8,
-	16,  //GE_TFMT_CLUT16,
-	32,  //GE_TFMT_CLUT32,
-	4,   //GE_TFMT_DXT1,
-	8,   //GE_TFMT_DXT3,
-	8,   //GE_TFMT_DXT5,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-};
-
-// Masks to downalign bufw to 16 bytes.
-static const u32 alignMask16[16] = {
-	~(((8 * 16) / 16) - 1),  //GE_TFMT_5650,
-	~(((8 * 16) / 16) - 1),  //GE_TFMT_5551,
-	~(((8 * 16) / 16) - 1),  //GE_TFMT_4444,
-	~(((8 * 16) / 32) - 1),  //GE_TFMT_8888,
-	~(((8 * 16) / 4) - 1),   //GE_TFMT_CLUT4,
-	~(((8 * 16) / 8) - 1),   //GE_TFMT_CLUT8,
-	~(((8 * 16) / 16) - 1),  //GE_TFMT_CLUT16,
-	~(((8 * 16) / 32) - 1),  //GE_TFMT_CLUT32,
-	~(((8 * 16) / 4) - 1),   //GE_TFMT_DXT1,
-	~(((8 * 16) / 8) - 1),   //GE_TFMT_DXT3,
-	~(((8 * 16) / 8) - 1),   //GE_TFMT_DXT5,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-	0,   // INVALID,
-};
-
 static inline u32 MiniHash(const u32 *ptr) {
 	return ptr[0];
 }
@@ -851,7 +806,7 @@ static inline u32 QuickClutHash(const u8 *clut, u32 bytes) {
 }
 
 static inline u32 QuickTexHash(u32 addr, int bufw, int w, int h, GETextureFormat format) {
-	const u32 sizeInRAM = (bitsPerPixel[format] * bufw * h) / 8;
+	const u32 sizeInRAM = (textureBitsPerPixel[format] * bufw * h) / 8;
 	const u32 *checkp = (const u32 *) Memory::GetPointer(addr);
 	u32 check = 0;
 
@@ -1055,10 +1010,10 @@ void TextureCache::SetTexture() {
 	} else {
 		cluthash = 0;
 	}
-
+	
+	int bufw = GetTextureBufw(0, texaddr, format);
 	int w = gstate.getTextureWidth(0);
 	int h = gstate.getTextureHeight(0);
-	int bufw = GetLevelBufw(0, texaddr);
 	int maxLevel = ((gstate.texmode >> 16) & 0x7);
 
 	u32 texhash = MiniHash((const u32 *)Memory::GetPointer(texaddr));
@@ -1211,7 +1166,7 @@ void TextureCache::SetTexture() {
 
 	// This would overestimate the size in many case so we underestimate instead
 	// to avoid excessive clearing caused by cache invalidations.
-	entry->sizeInRAM = (bitsPerPixel[format] * bufw * h / 2) / 8;
+	entry->sizeInRAM = (textureBitsPerPixel[format] * bufw * h / 2) / 8;
 
 	entry->fullhash = fullhash == 0 ? QuickTexHash(texaddr, bufw, w, h, format) : fullhash;
 	entry->cluthash = cluthash;
@@ -1308,12 +1263,7 @@ void *TextureCache::DecodeTextureLevel(GETextureFormat format, GEPaletteFormat c
 
 	u32 texaddr = (gstate.texaddr[level] & 0xFFFFF0) | ((gstate.texbufwidth[level] << 8) & 0x0F000000);
 
-	int bufw = GetLevelBufw(level, texaddr) & alignMask16[format];
+	int bufw = GetTextureBufw(level, texaddr, format);
-	if (bufw == 0) {
-		// If it's less than 16 bytes, use 16 bytes.
-		bufw = (8 * 16) / bitsPerPixel[format];
-	}
-
 	int w = gstate.getTextureWidth(level);
 	int h = gstate.getTextureHeight(level);
 	const u8 *texptr = Memory::GetPointer(texaddr);
@@ -1678,8 +1628,7 @@ bool TextureCache::DecodeTexture(u8* output, GPUgstate state)
 	GEPaletteFormat clutformat = gstate.getClutPaletteFormat();
 	u8 level = 0;
 
-	int bufw = GetLevelBufw(level, texaddr);
+	int bufw = GetTextureBufw(level, texaddr, format);
-
 	int w = gstate.getTextureWidth(level);
 	int h = gstate.getTextureHeight(level);
 

GPU/GPU.vcxproj

@@ -193,6 +193,7 @@
     <ClInclude Include="Software\Rasterizer.h" />
     <ClInclude Include="Software\SoftGpu.h" />
     <ClInclude Include="Software\TransformUnit.h" />
+    <ClInclude Include="Common\TextureDecoder.h" />
   </ItemGroup>
   <ItemGroup>
     <ClCompile Include="..\ext\xbrz\xbrz.cpp" />
@@ -243,6 +244,7 @@
     <ClCompile Include="Software\Rasterizer.cpp" />
     <ClCompile Include="Software\SoftGpu.cpp" />
     <ClCompile Include="Software\TransformUnit.cpp" />
+    <ClCompile Include="Common\TextureDecoder.cpp" />
   </ItemGroup>
   <ItemGroup>
     <ProjectReference Include="..\Common\Common.vcxproj">

GPU/GPU.vcxproj.filters

@@ -138,6 +138,9 @@
     <ClInclude Include="Directx9\helper\global.h">
       <Filter>DirectX9\helper</Filter>
     </ClInclude>
+    <ClInclude Include="Common\TextureDecoder.h">
+      <Filter>Common</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <ClCompile Include="Math3D.cpp">
@@ -254,6 +257,9 @@
     <ClCompile Include="Directx9\helper\global.cpp">
       <Filter>DirectX9\helper</Filter>
     </ClCompile>
+    <ClCompile Include="Common\TextureDecoder.cpp">
+      <Filter>Common</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <None Include="CMakeLists.txt" />

GPU/Software/Rasterizer.cpp

@@ -19,6 +19,7 @@
 #include "Core/Reporting.h"
 #include "GPU/GPUState.h"
 
+#include "GPU/Common/TextureDecoder.h"
 #include "GPU/Software/SoftGpu.h"
 #include "GPU/Software/Rasterizer.h"
 #include "GPU/Software/Colors.h"
@@ -171,8 +172,7 @@ static inline u32 SampleNearest(int level, unsigned int u, unsigned int v)
 	u32 texaddr = (gstate.texaddr[level] & 0xFFFFF0) | ((gstate.texbufwidth[level] << 8) & 0x0F000000);
 	u8* srcptr = (u8*)Memory::GetPointer(texaddr); // TODO: not sure if this is the right place to load from...?
 
-	// Special rules for kernel textures (PPGe), TODO: Verify!
+	int texbufwidth = GetTextureBufw(level, texaddr, texfmt);
-	int texbufwidth = (texaddr < PSP_GetUserMemoryBase()) ? gstate.texbufwidth[level] & 0x1FFF : gstate.texbufwidth[level] & 0x7FF;
 
 	// TODO: Should probably check if textures are aligned properly...
 

android/jni/Android.mk

@@ -194,6 +194,7 @@ LOCAL_SRC_FILES := \
   $(SRC)/GPU/GeDisasm.cpp \
   $(SRC)/GPU/Common/IndexGenerator.cpp.arm \
   $(SRC)/GPU/Common/VertexDecoderCommon.cpp.arm \
+  $(SRC)/GPU/Common/TextureDecoder.cpp \
   $(SRC)/GPU/GLES/Framebuffer.cpp \
   $(SRC)/GPU/GLES/GLES_GPU.cpp.arm \
   $(SRC)/GPU/GLES/TextureCache.cpp.arm \